MoEActok：面向视觉-语言-动作模型的基于混合专家的动作分词器
MoEActok: A MoE-based Action Tokenizer for Vision-Language-Action Models
对做视觉-语言-动作模型的研究者，这篇提供了用聚类驱动的 MoE 量化器把连续动作分块离散化成技能感知 token 的方案，可直接替换 VLA 中单一动作分词器，缓解多技能轨迹上的优化折中，并给出仿真与真机零样本迁移的评测设定。
原文依据 a novel action tokenizer that employs a mixture-of-experts (MoE) quantizer to produce skill-aware discrete representations for VLA models
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期关于视觉-语言-动作（VLA）模型的工作在探索动作分词器方面取得了很大进展，这类分词器将连续控制信号转换为离散 token，以对齐 LLM/VLM 的训练范式。这些方法通常在整个操作轨迹上训练单个分词器，而轨迹往往包含多种不同的技能，因而带来困难的优化折中。为解决该问题，我们提出 MoEActok，一种新的动作分词器，它采用混合专家（MoE）量化器为 VLA 模型产生技能感知的离散表示。MoEActok 采用聚类驱动的 MoE VQ-VAE 机制，其中每个专家专门负责某一特定技能。关键组件包括：(a) 一种动作-技能解耦策略，使用 k-means 聚类对动作分块进行分组，使具有相似技能的簇相互对齐；(b) 一种技能感知训练范式，为 VLA 模型增强技能条件上下文，改善技能 grounding；(c) 一个适配器，将共享编码器表示投影到技能特定的潜变量空间以进行专门量化，随后将异构的量化表示协调回统一空间，以便共享解码器进行连贯重建。我们在 RoboTwin 和 Simpler-Env 仿真器中，将基于 MoEActok 的 VLA 模型与多个先前的动作分词器基线进行比较，并进一步在三个真实世界任务上评估零样本迁移。在仿真与真实世界设定下，基于 MoEActok 的 VLA 均显著优于现有的离散分词方法。
英文原文摘要 1578 字符
Recent works on vision-language-action (VLA) models have made great progress in exploring action tokenizers that convert continuous control signals into discrete tokens to align with LLM/VLM training paradigms.These approaches typically train a single tokenizer over entire manipulation trajectories, which often comprise multiple distinct skills and thus pose a challenging optimization trade-off.To address this issue, we introduce MoEActok, a novel action tokenizer that employs a mixture-of-experts (MoE) quantizer to produce skill-aware discrete representations for VLA models. MoEActok utilizes a clustering-driven MoE VQ-VAE mechanism in which each expert specializes in a particular skill.The key components are: (a) an action-skill decoupling strategy that uses k-means clustering to group action chunks, aligning clusters having similar skills; (b) a skill-aware training paradigm that augments VLA models with skill-conditioned context, improving skill grounding; and (c) an adapter that projects shared encoder representations into skill-specific latent spaces for specialized quantization, and subsequently harmonize the heterogeneous quantized representations back into a unified space for coherent reconstruction by the shared decoder.We evaluate MoEActok-based VLA models against multiple prior action tokenizer baselines in the RoboTwin and Simpler-Env simulators, and further assess zero-shot transfer on three real-world tasks. Across both simulated and real-world settings, MoEActok-based VLA substantially outperforms existing discrete tokenization methods.